Abstract

The AI Agent Paid Twice: Intent-to-Outcome Verification for AI Shopping Agents Across Sellers, Payees, Hidden Fees, Subscriptions, Retries and Delivery

AI shopping agents now discover products, build carts, accept terms, pay and track delivery across marketplaces, merchants, payment providers and carriers, often with the user seeing only the first step. Every step produces a signal that is easy to over-read. A recommendation is not consent. A consent screen may hide the shipping charge or the renewal inside a free trial. A valid payment credential is not authority for this product from this seller. A marketplace listing may route money to a different beneficiary. A captured charge is not an accepted order, an accepted order is not the right item delivered, and a timeout followed by a retry can charge the user twice.

This disclosure describes a vendor-neutral, non-actuating method for checking that an autonomous purchase stays consistent from intent to outcome. Mandatory requirements are separated from preferences. Product, variant, quantity, actual seller, merchant of record and payment beneficiary are bound to the user’s approval. The complete obligation is recomputed in integer minor currency units (unit price × quantity + tax + shipping + fees − discounts) and compared with the checkout total, the approved maximum and the remaining shared budget. A purchase is admitted only when the exact pending operation matches the approved checkout revision under current, unrevoked and unused authority. Recurring and trial terms require their own separate authority. An unknown remote outcome blocks any retry until non-acceptance of the earlier attempt is proven. Payment, order, fulfilment and refund outcomes are kept as separate evidence, and three verdicts (preflight, outcome and overall) are computed in three-valued logic so that a contradiction prevails over unknowns and an unknown outcome is never reported as success.

The document provides a 60-stage, six-phase commerce workflow and a separate 52-stage payment-specific workflow, each with flowcharts, 72 plus 64 technical invariants, twelve technical variations and 60 engineering checkpoints. It compares the method with agentic checkout protocols, network and wallet tokens, 3-D Secure and strong customer authentication, payment idempotency keys, authorization and capture separation, spending controls, marketplace and payee verification, and subscription regulation. A non-actuating reference evaluator in Python, JavaScript and Go passes 89 shared synthetic test vectors. It never orders, pays or handles credentials; the laboratory is synthetic, self-authored and not peer reviewed.

Industry relevance: The problem addressed applies to AI platforms building agentic checkout (for example OpenAI’s Instant Checkout and Agentic Commerce Protocol with Stripe, Google’s Agent Payments Protocol, Amazon’s shopping agents, and MCP-based commerce tools), to payment networks and processors (for example Visa Intelligent Commerce, Mastercard Agent Pay, Stripe, PayPal, Adyen and Shopify), and to marketplaces with third-party sellers. It is also relevant to PSD2/PSD3, Verification of Payee, the Digital Services Act and subscription and fee regulation. This industry-relevance statement is AI-generated and does not assert that any named company uses, implements or infringes the disclosed method. Readers are strongly advised to read the original PCT filings before drawing any conclusion.

This disclosure is based on the Mothership application PCT/IB2026/055615, filed 4 June 2026 and publicly disclosed 16 July 2026, which claims priority from PCT/IB2026/053385, filed April 2026 and publicly disclosed 4 June 2026. The author is open to strategic collaboration, licensing or acquihire discussions.

Keywords: agentic commerce; AI shopping agent; autonomous purchasing; agent-initiated payment; purchase intent binding; user mandate; checkout integrity; merchant of record; payee verification; marketplace seller binding; total obligation recomputation; hidden fees; drip pricing; free trial and auto-renewal; payment authorization and capture; idempotency key; duplicate charge prevention; unknown payment outcome; retry safety; shared budget; multi-agent spending control; order and fulfilment verification; refund reconciliation; three-valued logic; AI agent safety.

Suggested CPC classes (for search): G06Q 20/40; G06Q 20/4016; G06Q 20/12; G06Q 30/0633; G06Q 30/0635; G06Q 30/0641; G06Q 20/389; G06N 5/04.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License

Download

Share

COinS